← back to paper
arxiv: 2602.09297 · 2 revisions
Laplacian Heads Improve Transformers by Smoothing Token Representations